The Only Way to Fight Deepfakes Is by Making Deepfakes
This article explores the emerging cottage industry of deepfake detection startups, such as Reality Defender, Pindrop, and GetReal, which utilize artificial intelligence to combat AI-generated manipulated media. Valued at approximately $5.5 billion as of 2023, this sector primarily addresses corporate fraud, though deepfakes are also used for harassment, political interference, and scams. The author describes a personal experiment where a voice clone generated by Reality Defender attempted to mimic their voice during a call with parents. While the attempt failed due to audio quality issues and robotic tones, it illustrates the core methodology of detection firms: training AI models using a student-teacher paradigm on both real and fake data. The text highlights the proliferation of consumer-grade AI tools that make creating fake audio and video frictionless, citing examples like nonconsensual sexual deepfakes and political robocalls. Ultimately, the piece argues that effectively identifying synthetic media requires the capability to generate it, positioning AI-driven detection as a necessary countermeasure to the industrial-scale rise of deepfake fraud.
Wire timeline
The Only Way to Fight Deepfakes Is by Making Deepfakes
This article explores the emerging cottage industry of deepfake detection startups, such as Reality Defender, Pindrop, and GetReal, which utilize artificial intelligence to combat AI-generated manipulated media. Valued at approximately $5.5 billion as of 2023, this sector primarily addresses corporate fraud, though deepfakes are also used for harassment, political interference, and scams. The author describes a personal experiment where a voice clone generated by Reality Defender attempted to mimic their voice during a call with parents. While the attempt failed due to audio quality issues and robotic tones, it illustrates the core methodology of detection firms: training AI models using a student-teacher paradigm on both real and fake data. The text highlights the proliferation of consumer-grade AI tools that make creating fake audio and video frictionless, citing examples like nonconsensual sexual deepfakes and political robocalls. Ultimately, the piece argues that effectively identifying synthetic media requires the capability to generate it, positioning AI-driven detection as a necessary countermeasure to the industrial-scale rise of deepfake fraud.
The Verge